TerraQ: Spatiotemporal Question-Answering on Satellite Image Archives
Sergios-Anestis Kefalidis, Konstantinos Plas, Manolis Koubarakis
TL;DR
The paper tackles making Earth Observation archives accessible through natural language by translating NL queries into SPARQL over a tailored geospatial KG. It presents TerraQ, a non-template spatiotemporal QA engine that interlinks geospatial features, administrative regions, and satellite image metadata to answer complex criteria. The system uses a four-stage NL-to-SPARQL pipeline with instance/concept mapping, spatial reasoning, and optional query refinement, plus offline materialization for performance. Compared to template-based approaches, TerraQ achieves higher accuracy on GeoQuestions1089 and runs on commodity hardware without relying on large LLMs.
Abstract
TerraQ is a spatiotemporal question-answering engine for satellite image archives. It is a natural language processing system that is built to process requests for satellite images satisfying certain criteria. The requests can refer to image metadata and entities from a specialized knowledge base (e.g., the Emilia-Romagna region). With it, users can make requests like "Give me a hundred images of rivers near ports in France, with less than 20% snow coverage and more than 10% cloud coverage", thus making Earth Observation data more easily accessible, in-line with the current landscape of digital assistants.